Compensation Methods for Industrial Robotics Under Varying Payloads with Deep Reinforcement Learning

Abstract

Due to the weak rigidity of an industrial robot, its end effector usually has poor absolute positioning accuracy, especially under varying payloads. Such situation is common in scenarios of handling, machining and tool changing. Conventional off-line calibration or compensation methods can only eliminate systematic errors, while such methods are invalid to the dynamic errors brought by varying payloads. This paper proposes a deep reinforcement learning(DRL) approach to solve the problem of dynamic errors, in consideration of external payloads changed manually. An online full closed loop system is established to verify the proposed method, which consists of a KUKA robot KR6, a Leica laser tracker, and a BECKHOFF PLC controller. The robot and the laser tracker work as the slavers of the master PLC controller, in between the communication is accomplished using EtherCAT. Logically, the robot is controlled by mxAutomation and the laser tracker is connected to an embedded EtherCAT slave card. Experiments on the robot demonstrate the effectiveness of the proposed DRL methods. The changed payloads range from 1.177Kg to 4.179 Kg, while the position accuracy of the robot can be maintained no more than 0.4mm by the DRL algorithm.